Current silence direction is quite strict so it starts transcription after a short pause in speech. Often these kinds of tools allow a user parameter to enter to allow a longer pause before transcription is started. My suggestion is to allow this in settings.